3 days of architecture, 3 days of implementation: Longbridge's Native Shell hit 1,700 stars in three weeks, riding the author's name

On Oct 19, a well-known engineer in the Chinese developer community posted: he and a colleague spent 6 days building a Rust cross-platform UI framework, where you write once in TSX and it renders as native controls on five operating systems. The repo gained 213 stars that day. Three weeks later it had over 1,700 stars, with a second wave. Below: what it is, where the stars came from, and which parts of a project like this you can copy and which you can't.

A tweet saying "this is how we work, 1000x faster"

On Oct 2026 19, huacnlee (Jason Lee, VP of technology at Longbridge Securities and founder of Ruby China) posted this on X:

@sunli829 and I spent 3 days discussing the architecture for this Native Shell and 3 days implementing it. It now runs on iOS, Android, macOS, Windows and Linux. Written in TSX, fully native rendering, and dynamic delivery. This is how we work, 1000x faster! This week we start using it in the Longbridge mobile app, mainly to support AI-generated modules on the fly.

That tweet had about 29k views and 220 likes by 11-06. In the replies he added that the two of them had burned through 2 Claude Premium and 2 Claude Max (20x) quotas in those few days. That day longbridge/nsh gained 213 stars and reached #8 on Trendshift's Rust daily chart (Trendshift, 10-19).

The repo was created on 10-14, with v10-17 released on 0.1.0.

What it is

Native Shell (NSH for short) is a cross-platform UI framework from Longbridge, with the engine written in Rust. You can write in Rust, or in TSX with Tailwind. One codebase renders as the system's own native controls on iOS, Android, macOS, Windows and Linux, as HTML on the web, and there are also GPU custom-drawing and terminal backends.

It also has a capability few rivals offer: at runtime it delivers signed TSX modules and runs them in a sandbox the host controls. Longbridge wanted this so AI-generated interfaces could appear live inside the app. Apache-2.0 license.

This space is already crowded

(GitHub,11-06)

NSH sits at the intersection of these: a Rust engine, native controls, TSX support, one codebase that covers even the terminal, and runtime delivery. But on tech alone, ByteDance's Lynx and Snap's Valdi already have very similar capabilities, backed by big companies. The differentiation is a ticket to be taken seriously, not the reason it's growing.

Where the stars came from: two waves, both on X

Wave one was the author himself. 10-18 +92, 10-19 +213 (Trendshift daily data), which lines up with that tweet. After that, dozens a day. On HN, 10-19 through 11-06 had 5 submissions in total, each scoring 1 to 3 points, and none of them took off.

An older project was funneling traffic. The author's GPUI Kit has 19.6k stars, and its README has a dedicated section, "Related project: Native Shell," and the author's GitHub profile also links to NSH. Its existing followers have been flowing into the new project steadily.

Wave two landed between 11-04 and 11-06, with +209, +291, and +274 over three days (Trendshift), higher than wave one. What happened in those days:

• 11-04, v0.4.0 was released. The English launch tweet got about 9,700 views and listed a string of screenshot-friendly new features: rich-text table editing, tree and heatmap charts, and shader effects.
• Early on 11-06, EGOIST (@localhost_5173, a well-known front-end developer) posted "You should probably check out nsh.dev," with "native controls, not self-drawn, works on mobile and desktop, written in Rust but you can also write it in TSX," about 11k views and 97 likes.
• The same day, @object_nullll posted a user story: "Yesterday I chose nsh to build the mobile version of eta. I set a strict condition: give it one evening, and if it doesn't work, switch to Expo. It was 10 p.m., I'd finished the design mockups, gave it a rough brief, and went off to kick off the task. When I'd just tidied up and was about to sleep, I found it could already collect the harvest."
• In 11-06 day, the author posted over a dozen times: terminal support done, system tray done, Longbridge's internal support system had switched from web to desktop, mobile, and Web clients all using NSH, and "30 to 50 PRs merged almost every day".

No single big-account tweet clearly explains the two-to-three-hundred-a-day volume; it looks more like these things stacking up, though there may be reposts I didn't find. Several technical blogs in Chinese, Korean, Japanese, and English (lijianfei.com, blog.mushroom.cv, geul.io, johan.se, etc.) amplified it, with unclear reach.

By late 11-06, GitHub showed 1,783 stars and 107 forks. 22 days saw 808 PRs merged, and 5 releases shipped in three weeks.

Why this one

• The author brings built-in credibility. huacnlee is well known among Chinese developers, and collaborator sunli829 is a known Rust developer (per public information, reportedly the author of async-graphql; not fully verified). When the new project launched, a crowd who already trusted them was waiting.
• The narrative rides on "AI writes code." It puts AI-friendliness front and center: TSX + Tailwind, which models know best; the repo includes llms.txt, a skills directory, AGENTS.md / CLAUDE.md; and the runtime can deliver AI-generated modules. User feedback lands on this point too, with one person saying "GPUI feels unfamiliar to AI and the output is garbage... want to try nsh."
• "Shipped in 6 days, burned through several Claude Max plans" is the story itself. In vibe coding discussions, that line travels further than any technical spec.
• Big companies are actually using it. Longbridge's mobile app is being integrated, and the internal customer service system has been rewritten, which suggests it isn't a toy (how deep the integration goes is only the author's claim).
• Every release has something demo-able, and each new capability is a reason to post.

Others are already following

oneo-me/nsgo, created one week after NSH, has about 10-21 stars as of 11-06 196. Its README is blunt: "NSGo follows the design of Native Shell (NSH)... It exists for teams that want NSH's model without writing Rust." It is a Go port of NSH's design, also supporting TSX + Tailwind, and doesn't need Node. It launched just one week after NSH, and has about a ninth of its stars.

There are also small apps the author and community have built with NSH (disktree, sysmon, stoatmd, tailpocket, etc.), all still tiny.

If it were you

Verdict: C. You can learn something here, but following it isn't worth much. Of NSH's 1,700 stars, the biggest chunks come from the author's name, the roughly 20k existing followers of GPUI Kit, and Longbridge as a real customer. You can't get any of those three. nsgo is a ready-made counterexample: same design, different language, up and running within a week, and 196 stars by week two. And UI frameworks are the heaviest, most crowded category of open source, with Flutter, React Native, and Lynx already present.

Who it suits: developers who already have some name recognition in a language community and can post something that people actually see; people who want to claim the "first" spot in a newly emerging ecosystem.

Who it doesn't suit: people who want to build a brand-new cross-platform UI framework from scratch and hope to get stars on technical merit alone; people with no posting channel at all.

How to do it, ranked by how worthwhile:

1. Be the "first" in the NSH ecosystem: a mature component library or full-page templates (chat, market quotes, settings pages), a starter and skills pack for "build an NSH app with AI in one evening," or an awesome-nsh list. The ecosystem has only just started, and these spots are still empty. The author is still posting daily, so if you do it well, he might repost you. The star ceiling is a few hundred, not thousands.
2. Copy its launch method for your own project: publish "built with AI in a few days" along with the token bill as the launch story; make the repo AI-friendly (llms.txt, skills, the patterns models know best) so the "AI built an app in one evening" user story spreads for you; add a link to the new project in the README of an older project; pair every release with a new capability you can screenshot.
3. Clone it in another language: nsgo has already claimed Go. Other languages (like Kotlin, Swift, or Zig) are possible, but given nsgo's numbers, set your expectations lower.

How to get users: post in the Chinese developer circle on X. Open with how many days you spent with AI and how much you burned through; attach a screen recording of it running on several platforms at once; @ the original author. Don't count on HN: NSH itself failed to take off there in five attempts.

When to move: the window for this niche is roughly before the end of 2026 (my estimate). NSH is currently merging 30 to 50 PRs a day, and the official team will likely fill in the component library and templates themselves.

How to tell you've misjudged: ① NSH takes off on HN or with big English accounts and stars climb past 10–20k; then the niche is worth far more than estimated here, and you should move faster. ② Conversely, if daily new stars fall back to single digits after mid-November and nobody outside Longbridge is really building apps with it, this wave was mostly the author effect, and the ecosystem isn't worth betting on. ③ If the official team ships its own component library and AI starter within a month or two, the first-mover path is closed.

Data and sources

GitHub: the longbridge/nsh repo, releases, contributors, and daily stars (10-18 to 11-06); the README for longbridge/gpui-kit; the README for oneo-me/nsgo. Trendshift: daily stars and the Rust chart at trendshift.io/repositories/309265. X: @huacnlee's 10-19 launch tweet, the 11-04 v0.4.0 release tweet and 11-06 thread series; posts by @localhost_5173 and @object_nullll 11-06 (view counts as 11-06). HN Algolia search. Blogs: lijianfei.com, blog.mushroom.cv, aisidaily.com, geul.io, johan.se. Anywhere marked "estimate" or "unverified" means exactly that.